How bad are things like AI searches for blogs and the web?
Really bad, it turns out. Even worse than I'd assumed. The drop in referrals when AI search is engaged is apparently *96 percent*.
https://www.forbes.com/sites/rashishrivastava/2025/03/03/openai-perplexity-ai-search-traffic-report/
That's scary for two reasons - firstly, it's bad for supporting anyone actually making content and will reduce content due to demand drops. Second, it implies a HUGE number of people are now just relying on AI slop to answer searches. Eek.
#Neovim here I come
For many reasons I have decided to make a real effort to use neovim to code with, as the standard IDE for my new project.
Big reason is that this is an infinity project. One I know never will be finished, just kept be improved, in code, as it is used.
So I really want to know and understand the code. Especially years later.
Thus the need to avoid any modern tempting shortcuts that especially #LLM's are about.
I think this will be fun too
#StarPro64 EIC7700X is the Hot New #RISCV SBC by @PINE64 ... Let's boot Linux and Apache #NuttX RTOS ... Maybe run #LLM on NPU on NuttX? (Large Language Model + Neural Processing Unit)
Re https://news.ycombinator.com/item?id=43154799 :
What a can of worms. It seems that 'reasoning' models are more prone to prompt injections than simpler ones.
Did anyone already do a comprehensive analysis?
#LLM vs Apache #NuttX RTOS: "Is it Safe to test this Pull Request on my computer?"
Source: https://gist.github.com/lupyuen/b9fc83a5f496d375b030c93c65271553
https://lornajane.net/posts/2025/from-http-to-openapi-with-optic
AI-powered search just got easier!
With Instant DeepSeek, OpenSearch lets you integrate DeepSeek LLMs with one API call—no manual setup, no complex orchestration.
One-click deployment
Automated pipelines
Faster LLM integration
Valentine's Day Giveaway
Win beta #RK3588AIModule7 A low - power #AI module compatible with Nvidia Jetson Nano. Great for high - performance computing and high - density servers.Get your hands on it early and share your thoughts!
To enter:
1、Repost and share your use case and desired features. https://docs.armsom.org/armsom-aim7
2、Subscribe to our crowdfunding:
https://www.crowdsupply.com/armsom/rk3588-ai-module7
Winners announced Feb 14th at 5 PM (Beijing Time).#opensource #linux #iot #hardware #machinelearning #llm #tech @itsfoss
ChatGPT is fairly convincing at creating code. But, like with everything you have to be vigilant on what it suggests you do. As a test I asked ChatGPT to "Write me an example C application using libcurl using secure HTTPS connection to fetch a file and save it locally. Provide instructions on how to create a test HTTPS server with self-signed certificate, and how to configure the server and the C client application for testing."
ChatGPT was fairly good here. It provided example code that didn't outright disable certificate validation, but rather uses the self-signed certificate as the CA store:
const char *cert_file = "./server.crt"; // Self-signed certificate
...
curl_easy_setopt(curl, CURLOPT_CAINFO, cert_file); // Verify server certificate
curl_easy_setopt(curl, CURLOPT_SSL_VERIFYPEER, 1L);
curl_easy_setopt(curl, CURLOPT_SSL_VERIFYHOST, 2L);
This is a very good idea, as blanket disabling security is a big nono. The deployment instructions were also quite nice, creating a self-signed certificate with openssl, and then setting up the test website with python3 http.server like this:
mkdir -p server
echo "This is a test file." > server/testfile.txt
python3 -m http.server 8443 --bind 127.0.0.1 --certfile server.crt --keyfile server.key
Looks pretty nice, right?
Except that this is totally hallucinated and even if it wasn't, it'd be totally insecure in a multiuser system anyway.
Python3 http.server doesn't allow you to pass certfile and keyfile like specified. But lets omit that small detail and assume it did. What would be the problem then?
You'd be sharing your whole work directory to everyone else on the same host. Anyone else on the same host could grab all your files with: wget --no-check-certificate -r https://127.0.0.1:8443
AI can be great, but never ever blindly trust the instructions provided by a LLM. They're not intelligent, but very good at pretending to be.
Meta torrented over 81.7TB of pirated books to train AI, authors say
https://arstechnica.com/tech-policy/2025/02/meta-torrented-over-81-7tb-of-pirated-books-to-train-ai-authors-say/
You nailed it @AuthorJMac !
Edit : Original (and complete) post on March 29, 2024
« You know what the biggest problem with pushing all-things-AI is? Wrong direction.
I want AI to do my laundry and dishes so that I can do art and writing, not for AI to do my art and writing so that I can do my laundry and dishes. »
Cf. https://indiepocalypse.social/@AuthorJMac/112178826967890119
2025 AIpub.social theme update
Looking for an instance dedicated to the exploration of artificial intelligence? Consider creating an account or migrating your existing account here!
We also have a Discord server home to over 600 members, you can find it here: https://discord.gg/DBvh2vPbzR
@geerlingguy I somehow approve of this content - it costs 5m to train an ai now, watch that cost plummet but there will be a big mkt. Let this year play out to see how quickly things evolve - it may be a lot faster than most anticipate? #free for commercial use #llm